When a batch fails in pharmaceutical manufacturing, the clock starts ticking. Regulatory requirements demand thorough investigation. Quality teams need answers. Production schedules hang in the balance.
In this post, we'll walk through how a pharmaceutical manufacturer transformed their root cause analysis process using Conduit.
The Challenge
Our customer operates a multi-site pharmaceutical manufacturing network. When quality events occur, investigators must piece together information from:
- Batch historians (OSIsoft PI)
- SCADA systems (Ignition)
- Quality management system (SAP QM)
- Environmental monitoring (Vaisala)
- Equipment logs (Various PLCs)
Before Conduit, a typical investigation looked like this:
- Quality engineer identifies the affected batch
- Requests data exports from each system (1-2 days)
- Receives Excel files from different teams (2-3 days)
- Manually correlates timestamps across systems (1-2 days)
- Identifies potential root causes (1 day)
- Requests additional data to confirm hypothesis (1-2 days)
- Documents findings (1 day)
Total time: 7-12 days per investigation
With hundreds of quality events annually, this consumed enormous resources and delayed production decisions.
The Solution
We deployed Conduit with adapters connecting to all critical systems:
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ OSIsoft PI │ │ Ignition │ │ SAP QM │
└──────┬──────┘ └──────┬──────┘ └──────┬──────┘
│ │ │
└──────────────────┼──────────────────┘
│
┌─────┴─────┐
│ Conduit │
└─────┬─────┘
│
┌───────────┼───────────┐
│ │ │
┌──────┴───┐ ┌─────┴────┐ ┌────┴─────┐
│ Vaisala │ │ PLC Logs │ │ MES Data │
└──────────┘ └──────────┘ └──────────┘
The semantic model maps batch identifiers across all systems, enabling instant correlation.
The Investigation: A Real Example
A batch failed viscosity testing. Here's how the investigation unfolded with Conduit:
Step 1: Initial Query
The quality engineer opens Conduit and types:
"Show me all process parameters for batch B-2024-1847 with any deviations from setpoints"
Within seconds, Conduit returns:
| Timestamp | System | Parameter | Setpoint | Actual | Deviation | |-----------|--------|-----------|----------|--------|-----------| | 14:23:15 | PI | Mixer Speed | 150 RPM | 142 RPM | -5.3% | | 14:45:00 | PI | Temperature | 65C | 67.2C | +3.4% | | 15:12:33 | Ignition | Tank Level | 850L | 823L | -3.2% |
The temperature deviation catches attention.
Step 2: Deep Dive
"Show me the temperature profile and cooling water valve position for reactor R-103 during batch B-2024-1847"
Conduit retrieves time-aligned data from PI (temperature) and the PLC historian (valve position):
Temperature peaked at 67.2C when cooling valve was at 78% open.
Normal operation: valve at 45-55% achieves setpoint.
Something was restricting cooling capacity.
Step 3: Cross-System Correlation
"Were there any maintenance events or alarms for the cooling system serving reactor R-103 in the 48 hours before batch B-2024-1847?"
Conduit queries Ignition's alarm history and the maintenance management system:
| Time | Event | Source | |------|-------|--------| | -36 hours | Cooling tower fan VFD fault | Ignition | | -35 hours | Work order created: CT-103 VFD replacement | SAP PM | | -12 hours | Work order closed: CT-103 VFD replaced | SAP PM |
The VFD had been replaced, but...
Step 4: Environmental Context
"Show me ambient temperature and cooling tower performance for the 24 hours before and during batch B-2024-1847"
From the environmental monitoring system:
Ambient temperature during batch: 34C (95th percentile for the year)
Cooling tower approach: 8.2C (normal: 4-5C)
Root cause identified: The combination of the recent VFD replacement (requiring a break-in period) and unusually high ambient temperature reduced cooling capacity below what was needed.
Step 5: Documentation
Conduit generates a timestamped investigation report including all queries, results, and conclusions. Total investigation time: 47 minutes.
The Results
After six months with Conduit:
| Metric | Before | After | Improvement | |--------|--------|-------|-------------| | Average investigation time | 9 days | 2 hours | 98% | | Data requests per investigation | 8 | 0 | 100% | | Investigations requiring rework | 23% | 4% | 83% | | Quality engineer utilization | 60% on data gathering | 90% on analysis | 50% |
Key Success Factors
Several factors contributed to successful deployment:
1. Semantic Model Design
We invested time upfront defining the semantic model:
- Batch identifiers mapped across all systems
- Equipment hierarchy aligned with ISA-95
- Time zone handling standardized
2. Read-Only Access
Conduit connects read-only to source systems. This simplified IT security approval and eliminated concerns about impacting production systems.
3. User Training
Quality engineers received hands-on training with real investigation scenarios. The natural language interface meant the learning curve was measured in hours, not weeks.
4. Iterative Refinement
We started with the three most-used systems and expanded monthly. Each new data source unlocked additional investigation capabilities.
Conclusion
Root cause analysis is fundamentally a data correlation problem. When data is fragmented across systems, investigations are slow and often incomplete. Conduit's federated query engine and semantic model transform this from a multi-day ordeal into a rapid, thorough process.
The pharmaceutical industry's regulatory requirements make this particularly valuable, but the same approach applies to any manufacturing environment where understanding "what happened" requires data from multiple systems.
Interested in transforming your root cause analysis process? Request a demo to see Conduit in action with your data sources.